Bit depth and color questions...Please help!

G

grahamdunn

Active member
Stu Maschwitz talks a lot about converting to linear color space, working in it, and then converting back on export. I'm confused as to what the 8, 16, and 32 bit settings are for in AE and how they relate to linear work flow. Clearly there is more color data in 16 and 32 bit than 8 if you use the right files, and the glows, exposure, etc. react well, but I'm still confused.

1. Why use 16 as opposed to 32 all the time?

2. How does the 8,16,32 bit thing tie to 8 and 10 bit files? What about people working in 64 bit?

3. Where does a 10-bit file come from? Only high end digital and film scans?

4. How does this stuff relate to say, 4:4:4 color or 4:2:2 color? Is 4:4:4 10 bit and everything else 8?

5. Why use a linear work flow when you do effects?


If anyone knows the answer to any of these questions, it would be greatly appreciated. There's info out there on how to work in these color spaces etc. but I don't really know when or why. Thanks!
 
1. easier on cpu/graphics processor
2. i think you are thinking of 8 and 10 bit source video footage vs. graphic bit depth.
3. for the most part, i think yes
4. not sure on that one
5. not sure of that... fm the sounds of it, it allows you to work faster, than up your BD before export... sounds like when you offline edit footage

I might be off a little, but i'm sure someone will chime in who knows answers to all...
 
Thanks for your reply...That's really helpful. I was mixing up video bit depth and graphics bit depth it sounds like. After a little research I found some more info that might be helpful to some people:

8 bit color has 256 colors per color channel, while 10 bit has 1024 colors per channel, so basically just way more color info. It's like the difference between shooting JPEG stills and RAW stills, you have more info in the highlights and that type of stuff. 8 and 10 bit relates to film and video footage, not graphics or compositing I guess.

4:4:4 and 4:2:2 and all that is based on YCbCr color space which is like Luminance, Blue, and Red. Basically your eye sees green's luminance (brightness) way more than red and blue. 4:4:4 samples green at every pixel, as well as blue and red, whereas 4:2:2 samples green at every pixel and red and blue at every other pixel. Since we don't notice red and blue as much, they are sampled less, allowing for smaller files with a minimal noticeable compression in color.

source: http://www.microsoft.com/windows/windowsmedia/howto/articles/UnderstandingHDFormats.aspx

I'd still love to know about linear color space...

Thanks all!
 
The benefit of working linear (32 bit) is that blending modes and other compositing functions react as they would in the real world, preserving values over 1 without clipping them, working in photographic terms (i.e., 50% gray solid in multiply will lower the image one stop), etc...

I recently did a project in the linear color workflow described by Stu Maschwitz in his blog. Below are three images of the same frame in the three different color depths. Here are the details of the project:

2 seperate 3d renders, one for the foreground (tree and cliff) and one for the background (sky and clouds). A greenscreened person sitting down against the screen, using the matte for a track matte of a black solid. The 3d renders were rendered out of C4D using OpenEXR which is a floating point (32bit) format. Project color space is set to Linear sRGB 2.1 (D65) which I downloaded from ProLost with Linearize Working Space set to on.

The only blending modes used are Add and Multiply. The only effects used are Levels, Hue/Saturation, and Radial Blur on the vignette.

8 bit

16 bit

32 bit

It's hard to tell the difference between the 8 and the 16, but if you look in the shadows under the cliff you can see it there. Obviously the 32 is in a league of it's own, I think the difference is very telling in the sky, the whole feel changes in 32 bit and looks better, more natural. It's easier to download the pics and open them in a viewer so you can quickly cycle through them, the difference will really stand out there.
 
3. Where does a 10-bit file come from? Only high end digital and film scans?

Eastman Kodak developed the Cineon format in order to capture the full dynamic range of motion picture film with their scanners while keeping file size as small as possible. The reason it could do that is that it is a 10bit log format.

Nature loves log. Our vision and hearing react in a logarithmic manner. Photographic emulsions also respond the same way, which is why film can capture so much nuance in the highlight and shadow areas.

However it would be very cumbersome to try to perform digital image processing on a log file. Even Kodak's Cineon software converted data to (larger) linear format while performing an operation.

Hi Def cameras have achieved the spatial resolution of film, but unlike film, sensors respond to light in a linear manner that drops some of the tonal subtleties near the ends of the curve. That is what the new HD production cameras are working to overcome, and they're getting close.
 
oneinfiniteloop—Thanks for the comparisons and descriptions of how you got there. The difference between 8 and 32 bit is striking; the sky and tree colors look very vivid and the whole scene seems to fit together more nicely. Thanks! (Is your project finished? I'd love to see it!)

Glen—Thank you for the background information. At work we shoot 35 and there's a certain something about it that's hard to describe, I guess that's the log vs linear. It's always described as organic, and based on what you said, that seems like an accurate dsecription.
 
I dont think you really need to ask, he details the benefits on his blog. I think graham just wanted a simpler explanation of the process co sidering its a lot to divest at one time.

@graham, its pretty much done, I'll post it soon.
 
Devon—I never did ask Stu, although Rebel Cafe has a lot of good info, and Stu has a very in-depth tutorial about linear color space. Oneinfiniteloop is right, I just wanted a sort of general explanation of what all was going on...The information on specific workflows is definitely out there though.

I'm looking forward to seeing your 32 bit project, oneinfiniteloop!

Thanks everyone!
 
Makes sense, but I'm just saying that if you need it clarified, you know where to get the info from the source as opposed to random forum people.
 
DevonKrain said:
Makes sense, but I'm just saying that if you need it clarified, you know where to get the info from the source as opposed to random forum people.
Click to expand...

As altruistic as this might be, floating point color is not something Stu Maschwitz created. Most of his questions are general knowledge questions and he needed simple answers to them. Sometimes experts like Mr. Maschwitz explain things in a way that's not as easy for a novice to understand. Although Stu Maschwitz does explain things in simple ways, there are even times when the best of us could get lost in all this color space stuff, especially with something as new as a linear workflow in AE.
 
Positive, he and Brendan Bolles facilitated a way to cheat floating point color in After Effects via eLin, but 32bit color depth is more a concept rather than a creation by one person. While I'm not trying to discount Stu Maschwitz in any way, nor am I discounting the great value it would be to ask or search his forums, I'm just saying graham's question was a valid one and "random forum people" can come across as slightly insulting when his questions are being answered correctly, which gives him a more clear way to understand what he has read over at ProLost.

Either way, this back and forth is getting old, I've learned just about everything I know about linear float workflow from Stu Maschwitz, Brendan Bolles, and Mark Christiansen's After Effects Studio Techniques (and his blog), as well as much experimentation, etc. I'm more than happy to help and share this knowledge to get someone else on the right path, even thought I may be a "random forum person"...whatever that may be.
 
What this means, oneinfiniteloop, (ironic name considering the conversation) is that if one has the option of asking the AUTHOR of books and software based on the concept or asking a "random forum member" who has or has not read said book(s), I know whom I would ask.

You can take offense all you want. It seems you're just getting a tad defensive.

I merely asked if he had asked Stu. You dont need to get offended that I think there might be someone more qualified to answer questions on the subject. This isnt a competition between authors and readers.

And this: "Are you sure?" was a joke. Which you seemed to recognize yet answer anyway. I can only assume it was to display your knowledge.
 
And you sir are exactly why I don't like responding in these forums. I understand what you are saying, and I apologize if I read into your post wrong, but you still shouldn't discount a forum like DVXUser where many professionals, including myself, Stu Maschwitz, and many others, come to answer questions.

Either way...I have work to do...

@graham, sorry for getting off-topic. Below is a link to the project that the still was pulled from.

http://www.tigerlilymedia.com/dcp/dcp_identFin_v1r2_H264.mov
(2.1mb's)
 
You must log in or register to reply here.
Back
Top